← All guides

Is ChatGPT Confidential? What Happens to What You Type

Kiyomi · 6 min read

Short answer: not by default, and for anyone with a confidentiality duty that distinction matters more than the quality of the answer.

What "confidential" has to mean for you

For a normal person, "confidential" means a company promises not to post your chats publicly. For a lawyer, doctor, accountant, therapist, or minister, it means something stricter: the information cannot leave your control, cannot be reviewed by a third party's staff, and cannot become training data for a model that then answers someone else's question. Public AI tools were not built to that standard. They were built for scale.

The three things that actually happen to your input

  1. It is transmitted to a third party. Your text leaves your device and is processed on the provider's servers. That alone can be a problem if you are handling privileged or regulated information without a data agreement in place.
  2. It may be retained. Most providers keep conversation history to operate the product. Retention windows vary, and "we delete it eventually" is not the same as "it never persisted."
  3. It may train future models. This is the one professionals miss. On consumer tiers, inputs can be used to improve the model unless you find and change a setting, and even then you are trusting the toggle. Enterprise tiers usually promise not to train on your data, but they cost accordingly and still route your data through the provider.

Why the settings are not enough

Turning off training in a consumer account reduces the risk; it does not remove the architecture. The data still travels to a third party, still sits on their infrastructure, and is still subject to their access, their subpoenas, and their breaches. For a professional, the question is not "did I click the right toggle" but "can this information leave my control at all."

What confidentiality-safe AI looks like

  • Processed on infrastructure the AI provider actually controls, not rented anonymously behind an API you cannot inspect.
  • Core content never used to train any model, as a property of how it is built, not a checkbox.
  • No sale, no sharing, no third-party review of what you bring to it.

That is the entire reason private AI exists as a category. The answer quality can match public tools; the difference is where your words go.

The short version

ChatGPT and its peers are not confidential in the sense a regulated professional needs. The inputs travel to a third party, may be retained, and may train future models. If your work carries a duty of confidentiality, that architecture is the problem, and a private AI built to keep your content yours is the answer.

Private AI that was built for this

Kiyomi runs on Jah, private AI on infrastructure Kiyomi controls. Your core chats are processed privately, never sold or shared, and never used to train a model. It is the AI you can use when confidentiality is not optional.

Try Kiyomi free — then $25/mo for everything.